freebsd-skq/sys/dev/usb/if_aue.c

1552 lines
36 KiB
C
Raw Normal View History

This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
/*
Attempt to fix a problem with receiving packets on USB ethernet interfaces. Packets are received inside USB bulk transfer callbacks, which run at splusb() (actually splbio()). The packet input queues are meant to be manipulated at splimp(). However the locking apparently breaks down under certain circumstances and the input queues can get trampled. There's a similar problem with if_ppp, which is driven by hardware/tty interrupts from the serial driver, but which must also manipulate the packet input queues at splimp(). The fix there is to use a netisr, and that's the fix I used here. (I can hear you groaning back there. Hush up.) The usb_ethersubr module maintains a single queue of its own. When a packet is received in the USB callback routine, it's placed on this queue with usb_ether_input(). This routine also schedules a soft net interrupt with schednetisr(). The ISR routine then runs later, at splnet, outside of the USB callback/interrupt context, and passes the packet to ether_input(), hopefully in a safe manner. The reason this is implemented as a separate module is that there are a limited number of NETISRs that we can use, and snarfing one up for each driver that needs it is wasteful (there will be three once I get the CATC driver done). It also reduces code duplication to a certain small extent. Unfortunately, it also needs to be linked in with the usb.ko module in order for the USB ethernet drivers to share it. Also removed some uneeded includes from if_aue.c and if_kue.c Fix suggested by: peter Not rejected as a hairbrained idea by: n_hibma
2000-01-10 23:12:54 +00:00
* Copyright (c) 1997, 1998, 1999, 2000
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
* Bill Paul <wpaul@ee.columbia.edu>. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by Bill Paul.
* 4. Neither the name of the author nor the names of any co-contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
* THE POSSIBILITY OF SUCH DAMAGE.
*
* $FreeBSD$
*/
/*
* ADMtek AN986 Pegasus USB to ethernet driver. Datasheet is available
* from http://www.admtek.com.tw.
*
* Written by Bill Paul <wpaul@ee.columbia.edu>
* Electrical Engineering Department
* Columbia University, New York City
*/
/*
* The Pegasus chip uses four USB "endpoints" to provide 10/100 ethernet
* support: the control endpoint for reading/writing registers, burst
* read endpoint for packet reception, burst write for packet transmission
* and one for "interrupts." The chip uses the same RX filter scheme
* as the other ADMtek ethernet parts: one perfect filter entry for the
* the station address and a 64-bit multicast hash table. The chip supports
* both MII and HomePNA attachments.
*
* Since the maximum data transfer speed of USB is supposed to be 12Mbps,
* you're never really going to get 100Mbps speeds from this device. I
* think the idea is to allow the device to connect to 10 or 100Mbps
* networks, not necessarily to provide 100Mbps performance. Also, since
* the controller uses an external PHY chip, it's possible that board
* designers might simply choose a 10Mbps PHY.
*
* Registers are accessed using usbd_do_request(). Packet transfers are
* done using usbd_transfer() and friends.
*/
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sockio.h>
#include <sys/mbuf.h>
#include <sys/malloc.h>
#include <sys/kernel.h>
#include <sys/socket.h>
#include <net/if.h>
#include <net/if_arp.h>
#include <net/ethernet.h>
#include <net/if_dl.h>
#include <net/if_media.h>
#include <net/bpf.h>
#include <machine/clock.h> /* for DELAY */
#include <sys/bus.h>
#include <dev/usb/usb.h>
#include <dev/usb/usbdi.h>
#include <dev/usb/usbdi_util.h>
#include <dev/usb/usbdivar.h>
#include <dev/usb/usbdevs.h>
Attempt to fix a problem with receiving packets on USB ethernet interfaces. Packets are received inside USB bulk transfer callbacks, which run at splusb() (actually splbio()). The packet input queues are meant to be manipulated at splimp(). However the locking apparently breaks down under certain circumstances and the input queues can get trampled. There's a similar problem with if_ppp, which is driven by hardware/tty interrupts from the serial driver, but which must also manipulate the packet input queues at splimp(). The fix there is to use a netisr, and that's the fix I used here. (I can hear you groaning back there. Hush up.) The usb_ethersubr module maintains a single queue of its own. When a packet is received in the USB callback routine, it's placed on this queue with usb_ether_input(). This routine also schedules a soft net interrupt with schednetisr(). The ISR routine then runs later, at splnet, outside of the USB callback/interrupt context, and passes the packet to ether_input(), hopefully in a safe manner. The reason this is implemented as a separate module is that there are a limited number of NETISRs that we can use, and snarfing one up for each driver that needs it is wasteful (there will be three once I get the CATC driver done). It also reduces code duplication to a certain small extent. Unfortunately, it also needs to be linked in with the usb.ko module in order for the USB ethernet drivers to share it. Also removed some uneeded includes from if_aue.c and if_kue.c Fix suggested by: peter Not rejected as a hairbrained idea by: n_hibma
2000-01-10 23:12:54 +00:00
#include <dev/usb/usb_ethersubr.h>
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
#include <dev/mii/mii.h>
#include <dev/mii/miivar.h>
#include <dev/usb/if_auereg.h>
/* "controller miibus0" required. See GENERIC if you get errors here. */
#include "miibus_if.h"
#ifndef lint
static const char rcsid[] =
"$FreeBSD$";
#endif
/*
* Various supported device vendors/products.
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
*/
static struct aue_type aue_devs[] = {
{ USB_VENDOR_ADMTEK, USB_PRODUCT_ADMTEK_PEGASUS },
{ USB_VENDOR_BILLIONTON, USB_PRODUCT_BILLIONTON_USB100 },
{ USB_VENDOR_MELCO, USB_PRODUCT_MELCO_LUATX },
{ USB_VENDOR_DLINK, USB_PRODUCT_DLINK_DSB650TX },
{ USB_VENDOR_DLINK, USB_PRODUCT_DLINK_DSB650TX_PNA },
{ USB_VENDOR_SMC, USB_PRODUCT_SMC_2202USB },
{ USB_VENDOR_LINKSYS, USB_PRODUCT_LINKSYS_USB100TX },
{ 0, 0 }
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
};
static struct usb_qdat aue_qdat;
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
static int aue_match __P((device_t));
static int aue_attach __P((device_t));
static int aue_detach __P((device_t));
static int aue_tx_list_init __P((struct aue_softc *));
static int aue_rx_list_init __P((struct aue_softc *));
static int aue_newbuf __P((struct aue_softc *, struct aue_chain *,
struct mbuf *));
static int aue_encap __P((struct aue_softc *, struct mbuf *, int));
static void aue_intr __P((usbd_xfer_handle,
usbd_private_handle, usbd_status));
static void aue_rxeof __P((usbd_xfer_handle,
usbd_private_handle, usbd_status));
static void aue_txeof __P((usbd_xfer_handle,
usbd_private_handle, usbd_status));
static void aue_tick __P((void *));
static void aue_rxstart __P((struct ifnet *));
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
static void aue_start __P((struct ifnet *));
static int aue_ioctl __P((struct ifnet *, u_long, caddr_t));
static void aue_init __P((void *));
static void aue_stop __P((struct aue_softc *));
static void aue_watchdog __P((struct ifnet *));
static void aue_shutdown __P((device_t));
static int aue_ifmedia_upd __P((struct ifnet *));
static void aue_ifmedia_sts __P((struct ifnet *, struct ifmediareq *));
static void aue_eeprom_getword __P((struct aue_softc *, int, u_int16_t *));
static void aue_read_eeprom __P((struct aue_softc *, caddr_t, int,
int, int));
static int aue_miibus_readreg __P((device_t, int, int));
static int aue_miibus_writereg __P((device_t, int, int, int));
static void aue_miibus_statchg __P((device_t));
static void aue_setmulti __P((struct aue_softc *));
static u_int32_t aue_crc __P((caddr_t));
static void aue_reset __P((struct aue_softc *));
static int csr_read_1 __P((struct aue_softc *, int));
static int csr_write_1 __P((struct aue_softc *, int, int));
static int csr_read_2 __P((struct aue_softc *, int));
static int csr_write_2 __P((struct aue_softc *, int, int));
static device_method_t aue_methods[] = {
/* Device interface */
DEVMETHOD(device_probe, aue_match),
DEVMETHOD(device_attach, aue_attach),
DEVMETHOD(device_detach, aue_detach),
DEVMETHOD(device_shutdown, aue_shutdown),
/* bus interface */
DEVMETHOD(bus_print_child, bus_generic_print_child),
DEVMETHOD(bus_driver_added, bus_generic_driver_added),
/* MII interface */
DEVMETHOD(miibus_readreg, aue_miibus_readreg),
DEVMETHOD(miibus_writereg, aue_miibus_writereg),
DEVMETHOD(miibus_statchg, aue_miibus_statchg),
{ 0, 0 }
};
static driver_t aue_driver = {
"aue",
aue_methods,
sizeof(struct aue_softc)
};
static devclass_t aue_devclass;
DRIVER_MODULE(if_aue, uhub, aue_driver, aue_devclass, usbd_driver_load, 0);
DRIVER_MODULE(miibus, aue, miibus_driver, miibus_devclass, 0, 0);
#define AUE_SETBIT(sc, reg, x) \
csr_write_1(sc, reg, csr_read_1(sc, reg) | (x))
#define AUE_CLRBIT(sc, reg, x) \
csr_write_1(sc, reg, csr_read_1(sc, reg) & ~(x))
static int csr_read_1(sc, reg)
struct aue_softc *sc;
int reg;
{
usb_device_request_t req;
usbd_status err;
u_int8_t val = 0;
int s;
s = splusb();
req.bmRequestType = UT_READ_VENDOR_DEVICE;
req.bRequest = AUE_UR_READREG;
USETW(req.wValue, 0);
USETW(req.wIndex, reg);
USETW(req.wLength, 1);
err = usbd_do_request(sc->aue_udev, &req, &val);
splx(s);
if (err)
return(0);
return(val);
}
static int csr_read_2(sc, reg)
struct aue_softc *sc;
int reg;
{
usb_device_request_t req;
usbd_status err;
u_int16_t val = 0;
int s;
s = splusb();
req.bmRequestType = UT_READ_VENDOR_DEVICE;
req.bRequest = AUE_UR_READREG;
USETW(req.wValue, 0);
USETW(req.wIndex, reg);
USETW(req.wLength, 2);
err = usbd_do_request(sc->aue_udev, &req, &val);
splx(s);
if (err)
return(0);
return(val);
}
static int csr_write_1(sc, reg, val)
struct aue_softc *sc;
int reg, val;
{
usb_device_request_t req;
usbd_status err;
int s;
s = splusb();
req.bmRequestType = UT_WRITE_VENDOR_DEVICE;
req.bRequest = AUE_UR_WRITEREG;
USETW(req.wValue, val);
USETW(req.wIndex, reg);
USETW(req.wLength, 1);
err = usbd_do_request(sc->aue_udev, &req, &val);
splx(s);
if (err)
return(-1);
return(0);
}
static int csr_write_2(sc, reg, val)
struct aue_softc *sc;
int reg, val;
{
usb_device_request_t req;
usbd_status err;
int s;
s = splusb();
req.bmRequestType = UT_WRITE_VENDOR_DEVICE;
req.bRequest = AUE_UR_WRITEREG;
USETW(req.wValue, val);
USETW(req.wIndex, reg);
USETW(req.wLength, 2);
err = usbd_do_request(sc->aue_udev, &req, &val);
splx(s);
if (err)
return(-1);
return(0);
}
/*
* Read a word of data stored in the EEPROM at address 'addr.'
*/
static void aue_eeprom_getword(sc, addr, dest)
struct aue_softc *sc;
int addr;
u_int16_t *dest;
{
register int i;
u_int16_t word = 0;
csr_write_1(sc, AUE_EE_REG, addr);
csr_write_1(sc, AUE_EE_CTL, AUE_EECTL_READ);
for (i = 0; i < AUE_TIMEOUT; i++) {
if (csr_read_1(sc, AUE_EE_CTL) &
AUE_EECTL_DONE)
break;
}
if (i == AUE_TIMEOUT) {
printf("aue%d: EEPROM read timed out\n",
sc->aue_unit);
}
word = csr_read_2(sc, AUE_EE_DATA);
*dest = word;
return;
}
/*
* Read a sequence of words from the EEPROM.
*/
static void aue_read_eeprom(sc, dest, off, cnt, swap)
struct aue_softc *sc;
caddr_t dest;
int off;
int cnt;
int swap;
{
int i;
u_int16_t word = 0, *ptr;
for (i = 0; i < cnt; i++) {
aue_eeprom_getword(sc, off + i, &word);
ptr = (u_int16_t *)(dest + (i * 2));
if (swap)
*ptr = ntohs(word);
else
*ptr = word;
}
return;
}
static int aue_miibus_readreg(dev, phy, reg)
device_t dev;
int phy, reg;
{
struct aue_softc *sc;
int i;
u_int16_t val = 0;
sc = device_get_softc(dev);
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
/*
* The Am79C901 HomePNA PHY actually contains
* two transceivers: a 1Mbps HomePNA PHY and a
* 10Mbps full/half duplex ethernet PHY with
* NWAY autoneg. However in the ADMtek adapter,
* only the 1Mbps PHY is actually connected to
* anything, so we ignore the 10Mbps one. It
* happens to be configured for MII address 3,
* so we filter that out.
*/
if (sc->aue_info->aue_vid == USB_VENDOR_ADMTEK &&
sc->aue_info->aue_did == USB_PRODUCT_ADMTEK_PEGASUS) {
if (phy == 3)
return(0);
if (phy != 1)
return(0);
}
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
csr_write_1(sc, AUE_PHY_ADDR, phy);
csr_write_1(sc, AUE_PHY_CTL, reg|AUE_PHYCTL_READ);
for (i = 0; i < AUE_TIMEOUT; i++) {
if (csr_read_1(sc, AUE_PHY_CTL) &
AUE_PHYCTL_DONE)
break;
}
if (i == AUE_TIMEOUT) {
printf("aue%d: MII read timed out\n",
sc->aue_unit);
}
val = csr_read_2(sc, AUE_PHY_DATA);
return(val);
}
static int aue_miibus_writereg(dev, phy, reg, data)
device_t dev;
int phy, reg, data;
{
struct aue_softc *sc;
int i;
if (phy == 3)
return(0);
sc = device_get_softc(dev);
csr_write_2(sc, AUE_PHY_DATA, data);
csr_write_1(sc, AUE_PHY_ADDR, phy);
csr_write_1(sc, AUE_PHY_CTL, reg|AUE_PHYCTL_WRITE);
for (i = 0; i < AUE_TIMEOUT; i++) {
if (csr_read_1(sc, AUE_PHY_CTL) &
AUE_PHYCTL_DONE)
break;
}
if (i == AUE_TIMEOUT) {
printf("aue%d: MII read timed out\n",
sc->aue_unit);
}
return(0);
}
static void aue_miibus_statchg(dev)
device_t dev;
{
struct aue_softc *sc;
struct mii_data *mii;
sc = device_get_softc(dev);
mii = device_get_softc(sc->aue_miibus);
AUE_CLRBIT(sc, AUE_CTL0, AUE_CTL0_RX_ENB|AUE_CTL0_TX_ENB);
if (IFM_SUBTYPE(mii->mii_media_active) == IFM_100_TX) {
AUE_SETBIT(sc, AUE_CTL1, AUE_CTL1_SPEEDSEL);
} else {
AUE_CLRBIT(sc, AUE_CTL1, AUE_CTL1_SPEEDSEL);
}
if ((mii->mii_media_active & IFM_GMASK) == IFM_FDX) {
AUE_SETBIT(sc, AUE_CTL1, AUE_CTL1_DUPLEX);
} else {
AUE_CLRBIT(sc, AUE_CTL1, AUE_CTL1_DUPLEX);
}
AUE_SETBIT(sc, AUE_CTL0, AUE_CTL0_RX_ENB|AUE_CTL0_TX_ENB);
/*
* Set the LED modes on the LinkSys adapter.
* This turns on the 'dual link LED' bin in the auxmode
* register of the Broadcom PHY.
*/
if (sc->aue_info->aue_vid == USB_VENDOR_LINKSYS &&
sc->aue_info->aue_did == USB_PRODUCT_LINKSYS_USB100TX) {
u_int16_t auxmode;
auxmode = aue_miibus_readreg(dev, 0, 0x1b);
aue_miibus_writereg(dev, 0, 0x1b, auxmode | 0x04);
}
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
return;
}
#define AUE_POLY 0xEDB88320
#define AUE_BITS 6
static u_int32_t aue_crc(addr)
caddr_t addr;
{
u_int32_t idx, bit, data, crc;
/* Compute CRC for the address value. */
crc = 0xFFFFFFFF; /* initial value */
for (idx = 0; idx < 6; idx++) {
for (data = *addr++, bit = 0; bit < 8; bit++, data >>= 1)
crc = (crc >> 1) ^ (((crc ^ data) & 1) ? AUE_POLY : 0);
}
return (crc & ((1 << AUE_BITS) - 1));
}
static void aue_setmulti(sc)
struct aue_softc *sc;
{
struct ifnet *ifp;
struct ifmultiaddr *ifma;
u_int32_t h = 0, i;
ifp = &sc->arpcom.ac_if;
if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
AUE_SETBIT(sc, AUE_CTL0, AUE_CTL0_ALLMULTI);
return;
}
AUE_CLRBIT(sc, AUE_CTL0, AUE_CTL0_ALLMULTI);
/* first, zot all the existing hash bits */
for (i = 0; i < 8; i++)
csr_write_1(sc, AUE_MAR0 + i, 0);
/* now program new ones */
for (ifma = ifp->if_multiaddrs.lh_first; ifma != NULL;
ifma = ifma->ifma_link.le_next) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
h = aue_crc(LLADDR((struct sockaddr_dl *)ifma->ifma_addr));
AUE_SETBIT(sc, AUE_MAR + (h >> 3), 1 << (h & 0xF));
}
return;
}
static void aue_reset(sc)
struct aue_softc *sc;
{
register int i;
AUE_SETBIT(sc, AUE_CTL1, AUE_CTL1_RESETMAC);
for (i = 0; i < AUE_TIMEOUT; i++) {
if (!(csr_read_1(sc, AUE_CTL1) & AUE_CTL1_RESETMAC))
break;
}
if (i == AUE_TIMEOUT)
printf("aue%d: reset failed\n", sc->aue_unit);
/*
* The PHY(s) attached to the Pegasus chip may be held
* in reset until we flip on the GPIO outputs. Make sure
* to set the GPIO pins high so that the PHY(s) will
* be enabled.
*
* Note: We force all of the GPIO pins low first, *then*
* enable the ones we want.
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
*/
csr_write_1(sc, AUE_GPIO0, AUE_GPIO_OUT0|AUE_GPIO_SEL0);
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
csr_write_1(sc, AUE_GPIO0, AUE_GPIO_OUT0|AUE_GPIO_SEL0|AUE_GPIO_SEL1);
/* Grrr. LinkSys has to be different from everyone else. */
if (sc->aue_info->aue_vid == USB_VENDOR_LINKSYS &&
sc->aue_info->aue_did == USB_PRODUCT_LINKSYS_USB100TX) {
csr_write_1(sc, AUE_GPIO0, AUE_GPIO_SEL0|AUE_GPIO_SEL1);
csr_write_1(sc, AUE_GPIO0, AUE_GPIO_SEL0|AUE_GPIO_SEL1|
AUE_GPIO_OUT0);
}
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
/* Wait a little while for the chip to get its brains in order. */
DELAY(10000);
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
return;
}
/*
* Probe for a Pegasus chip.
*/
USB_MATCH(aue)
{
USB_MATCH_START(aue, uaa);
struct aue_type *t;
if (!uaa->iface)
return(UMATCH_NONE);
t = aue_devs;
while(t->aue_vid) {
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
if (uaa->vendor == t->aue_vid &&
uaa->product == t->aue_did) {
return(UMATCH_VENDOR_PRODUCT);
}
t++;
}
return(UMATCH_NONE);
}
/*
* Attach the interface. Allocate softc structures, do ifmedia
* setup and ethernet/BPF attach.
*/
USB_ATTACH(aue)
{
USB_ATTACH_START(aue, sc, uaa);
char devinfo[1024];
int s;
u_char eaddr[ETHER_ADDR_LEN];
struct ifnet *ifp;
usb_interface_descriptor_t *id;
usb_endpoint_descriptor_t *ed;
int i;
struct aue_type *t;
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
s = splimp();
bzero(sc, sizeof(struct aue_softc));
sc->aue_iface = uaa->iface;
sc->aue_udev = uaa->device;
sc->aue_unit = device_get_unit(self);
if (usbd_set_config_no(sc->aue_udev, AUE_CONFIG_NO, 0)) {
printf("aue%d: getting interface handle failed\n",
sc->aue_unit);
splx(s);
USB_ATTACH_ERROR_RETURN;
}
t = aue_devs;
while(t->aue_vid) {
if (uaa->vendor == t->aue_vid &&
uaa->product == t->aue_did) {
sc->aue_info = t;
break;
}
t++;
}
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
id = usbd_get_interface_descriptor(uaa->iface);
usbd_devinfo(uaa->device, 0, devinfo);
device_set_desc_copy(self, devinfo);
printf("%s: %s\n", USBDEVNAME(self), devinfo);
/* Find endpoints. */
for (i = 0; i < id->bNumEndpoints; i++) {
ed = usbd_interface2endpoint_descriptor(uaa->iface, i);
if (!ed) {
printf("aue%d: couldn't get ep %d\n",
sc->aue_unit, i);
splx(s);
USB_ATTACH_ERROR_RETURN;
}
if (UE_GET_DIR(ed->bEndpointAddress) == UE_DIR_IN &&
(ed->bmAttributes & UE_XFERTYPE) == UE_BULK) {
sc->aue_ed[AUE_ENDPT_RX] = ed->bEndpointAddress;
} else if (UE_GET_DIR(ed->bEndpointAddress) == UE_DIR_OUT &&
(ed->bmAttributes & UE_XFERTYPE) == UE_BULK) {
sc->aue_ed[AUE_ENDPT_TX] = ed->bEndpointAddress;
} else if (UE_GET_DIR(ed->bEndpointAddress) == UE_DIR_IN &&
(ed->bmAttributes & UE_XFERTYPE) == UE_INTERRUPT) {
sc->aue_ed[AUE_ENDPT_INTR] = ed->bEndpointAddress;
}
}
/* Reset the adapter. */
aue_reset(sc);
/*
* Get station address from the EEPROM.
*/
aue_read_eeprom(sc, (caddr_t)&eaddr, 0, 3, 0);
/*
* A Pegasus chip was detected. Inform the world.
*/
printf("aue%d: Ethernet address: %6D\n", sc->aue_unit, eaddr, ":");
bcopy(eaddr, (char *)&sc->arpcom.ac_enaddr, ETHER_ADDR_LEN);
ifp = &sc->arpcom.ac_if;
ifp->if_softc = sc;
ifp->if_unit = sc->aue_unit;
ifp->if_name = "aue";
ifp->if_mtu = ETHERMTU;
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_ioctl = aue_ioctl;
ifp->if_output = ether_output;
ifp->if_start = aue_start;
ifp->if_watchdog = aue_watchdog;
ifp->if_init = aue_init;
ifp->if_baudrate = 10000000;
ifp->if_snd.ifq_maxlen = IFQ_MAXLEN;
/*
* Do MII setup.
* NOTE: Doing this causes child devices to be attached to us,
* which we would normally disconnect at in the detach routine
* using device_delete_child(). However the USB code is set up
* such that when this driver is removed, all children devices
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
* are removed as well. In effect, the USB code ends up detaching
* all of our children for us, so we don't have to do is ourselves
* in aue_detach(). It's important to point this out since if
* we *do* try to detach the child devices ourselves, we will
* end up getting the children deleted twice, which will crash
* the system.
*/
if (mii_phy_probe(self, &sc->aue_miibus,
aue_ifmedia_upd, aue_ifmedia_sts)) {
printf("aue%d: MII without any PHY!\n", sc->aue_unit);
splx(s);
USB_ATTACH_ERROR_RETURN;
}
aue_qdat.ifp = ifp;
aue_qdat.if_rxstart = aue_rxstart;
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
/*
* Call MI attach routines.
*/
if_attach(ifp);
ether_ifattach(ifp);
callout_handle_init(&sc->aue_stat_ch);
bpfattach(ifp, DLT_EN10MB, sizeof(struct ether_header));
Attempt to fix a problem with receiving packets on USB ethernet interfaces. Packets are received inside USB bulk transfer callbacks, which run at splusb() (actually splbio()). The packet input queues are meant to be manipulated at splimp(). However the locking apparently breaks down under certain circumstances and the input queues can get trampled. There's a similar problem with if_ppp, which is driven by hardware/tty interrupts from the serial driver, but which must also manipulate the packet input queues at splimp(). The fix there is to use a netisr, and that's the fix I used here. (I can hear you groaning back there. Hush up.) The usb_ethersubr module maintains a single queue of its own. When a packet is received in the USB callback routine, it's placed on this queue with usb_ether_input(). This routine also schedules a soft net interrupt with schednetisr(). The ISR routine then runs later, at splnet, outside of the USB callback/interrupt context, and passes the packet to ether_input(), hopefully in a safe manner. The reason this is implemented as a separate module is that there are a limited number of NETISRs that we can use, and snarfing one up for each driver that needs it is wasteful (there will be three once I get the CATC driver done). It also reduces code duplication to a certain small extent. Unfortunately, it also needs to be linked in with the usb.ko module in order for the USB ethernet drivers to share it. Also removed some uneeded includes from if_aue.c and if_kue.c Fix suggested by: peter Not rejected as a hairbrained idea by: n_hibma
2000-01-10 23:12:54 +00:00
usb_register_netisr();
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
splx(s);
USB_ATTACH_SUCCESS_RETURN;
}
static int aue_detach(dev)
device_t dev;
{
struct aue_softc *sc;
struct ifnet *ifp;
int s;
s = splusb();
sc = device_get_softc(dev);
ifp = &sc->arpcom.ac_if;
untimeout(aue_tick, sc, sc->aue_stat_ch);
if_detach(ifp);
if (sc->aue_ep[AUE_ENDPT_TX] != NULL)
usbd_abort_pipe(sc->aue_ep[AUE_ENDPT_TX]);
if (sc->aue_ep[AUE_ENDPT_RX] != NULL)
usbd_abort_pipe(sc->aue_ep[AUE_ENDPT_RX]);
if (sc->aue_ep[AUE_ENDPT_INTR] != NULL)
usbd_abort_pipe(sc->aue_ep[AUE_ENDPT_INTR]);
splx(s);
return(0);
}
/*
* Initialize an RX descriptor and attach an MBUF cluster.
*/
static int aue_newbuf(sc, c, m)
struct aue_softc *sc;
struct aue_chain *c;
struct mbuf *m;
{
struct mbuf *m_new = NULL;
if (m == NULL) {
MGETHDR(m_new, M_DONTWAIT, MT_DATA);
if (m_new == NULL) {
printf("aue%d: no memory for rx list "
"-- packet dropped!\n", sc->aue_unit);
return(ENOBUFS);
}
MCLGET(m_new, M_DONTWAIT);
if (!(m_new->m_flags & M_EXT)) {
printf("aue%d: no memory for rx list "
"-- packet dropped!\n", sc->aue_unit);
m_freem(m_new);
return(ENOBUFS);
}
m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
} else {
m_new = m;
m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
m_new->m_data = m_new->m_ext.ext_buf;
}
m_adj(m_new, ETHER_ALIGN);
c->aue_mbuf = m_new;
return(0);
}
static int aue_rx_list_init(sc)
struct aue_softc *sc;
{
struct aue_cdata *cd;
struct aue_chain *c;
int i;
cd = &sc->aue_cdata;
for (i = 0; i < AUE_RX_LIST_CNT; i++) {
c = &cd->aue_rx_chain[i];
c->aue_sc = sc;
c->aue_idx = i;
c->aue_accum = 0;
if (aue_newbuf(sc, c, NULL) == ENOBUFS)
return(ENOBUFS);
if (c->aue_xfer == NULL) {
c->aue_xfer = usbd_alloc_xfer(sc->aue_udev);
if (c->aue_xfer == NULL)
return(ENOBUFS);
}
}
return(0);
}
static int aue_tx_list_init(sc)
struct aue_softc *sc;
{
struct aue_cdata *cd;
struct aue_chain *c;
int i;
cd = &sc->aue_cdata;
for (i = 0; i < AUE_TX_LIST_CNT; i++) {
c = &cd->aue_tx_chain[i];
c->aue_sc = sc;
c->aue_idx = i;
c->aue_mbuf = NULL;
if (c->aue_xfer == NULL) {
c->aue_xfer = usbd_alloc_xfer(sc->aue_udev);
if (c->aue_xfer == NULL)
return(ENOBUFS);
}
c->aue_buf = malloc(AUE_BUFSZ, M_USBDEV, M_NOWAIT);
if (c->aue_buf == NULL)
return(ENOBUFS);
}
return(0);
}
static void aue_intr(xfer, priv, status)
usbd_xfer_handle xfer;
usbd_private_handle priv;
usbd_status status;
{
struct aue_softc *sc;
struct ifnet *ifp;
struct aue_intrpkt *p;
int s;
s = splimp();
sc = priv;
ifp = &sc->arpcom.ac_if;
if (!(ifp->if_flags & IFF_RUNNING)) {
splx(s);
return;
}
if (status != USBD_NORMAL_COMPLETION) {
if (status == USBD_NOT_STARTED || status == USBD_CANCELLED) {
splx(s);
return;
}
printf("aue%d: usb error on intr: %s\n", sc->aue_unit,
usbd_errstr(status));
if (status == USBD_STALLED)
usbd_clear_endpoint_stall(sc->aue_ep[AUE_ENDPT_RX]);
splx(s);
return;
}
usbd_get_xfer_status(xfer, NULL, (void **)&p, NULL, NULL);
if (p->aue_txstat0)
ifp->if_oerrors++;
if (p->aue_txstat0 & (AUE_TXSTAT0_LATECOLL | AUE_TXSTAT0_EXCESSCOLL))
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
ifp->if_collisions++;
splx(s);
return;
}
static void aue_rxstart(ifp)
struct ifnet *ifp;
{
struct aue_softc *sc;
struct aue_chain *c;
sc = ifp->if_softc;
c = &sc->aue_cdata.aue_rx_chain[sc->aue_cdata.aue_rx_prod];
if (aue_newbuf(sc, c, NULL) == ENOBUFS) {
ifp->if_ierrors++;
return;
}
/* Setup new transfer. */
usbd_setup_xfer(c->aue_xfer, sc->aue_ep[AUE_ENDPT_RX],
c, mtod(c->aue_mbuf, char *), AUE_CUTOFF, USBD_SHORT_XFER_OK,
USBD_NO_TIMEOUT, aue_rxeof);
usbd_transfer(c->aue_xfer);
return;
}
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
/*
* A frame has been uploaded: pass the resulting mbuf chain up to
* the higher level protocols.
*
* Grrr. Receiving transfers larger than about 1152 bytes sometimes
* doesn't work. We get an incomplete frame. In order to avoid
* this, we queue up RX transfers that are shorter than a full sized
* frame. If the received frame is larger than our transfer size,
* we snag the rest of the data using a second transfer. Does this
* hurt performance? Yes. But after fighting with this stupid thing
* for three days, I'm willing to settle. I'd rather have reliable
* receive performance that fast but spotty performance.
*/
static void aue_rxeof(xfer, priv, status)
usbd_xfer_handle xfer;
usbd_private_handle priv;
usbd_status status;
{
struct aue_softc *sc;
struct aue_chain *c;
struct mbuf *m;
struct ifnet *ifp;
int total_len = 0;
struct aue_rxpkt r;
c = priv;
sc = c->aue_sc;
ifp = &sc->arpcom.ac_if;
Attempt to fix a problem with receiving packets on USB ethernet interfaces. Packets are received inside USB bulk transfer callbacks, which run at splusb() (actually splbio()). The packet input queues are meant to be manipulated at splimp(). However the locking apparently breaks down under certain circumstances and the input queues can get trampled. There's a similar problem with if_ppp, which is driven by hardware/tty interrupts from the serial driver, but which must also manipulate the packet input queues at splimp(). The fix there is to use a netisr, and that's the fix I used here. (I can hear you groaning back there. Hush up.) The usb_ethersubr module maintains a single queue of its own. When a packet is received in the USB callback routine, it's placed on this queue with usb_ether_input(). This routine also schedules a soft net interrupt with schednetisr(). The ISR routine then runs later, at splnet, outside of the USB callback/interrupt context, and passes the packet to ether_input(), hopefully in a safe manner. The reason this is implemented as a separate module is that there are a limited number of NETISRs that we can use, and snarfing one up for each driver that needs it is wasteful (there will be three once I get the CATC driver done). It also reduces code duplication to a certain small extent. Unfortunately, it also needs to be linked in with the usb.ko module in order for the USB ethernet drivers to share it. Also removed some uneeded includes from if_aue.c and if_kue.c Fix suggested by: peter Not rejected as a hairbrained idea by: n_hibma
2000-01-10 23:12:54 +00:00
if (!(ifp->if_flags & IFF_RUNNING))
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
return;
if (status != USBD_NORMAL_COMPLETION) {
Attempt to fix a problem with receiving packets on USB ethernet interfaces. Packets are received inside USB bulk transfer callbacks, which run at splusb() (actually splbio()). The packet input queues are meant to be manipulated at splimp(). However the locking apparently breaks down under certain circumstances and the input queues can get trampled. There's a similar problem with if_ppp, which is driven by hardware/tty interrupts from the serial driver, but which must also manipulate the packet input queues at splimp(). The fix there is to use a netisr, and that's the fix I used here. (I can hear you groaning back there. Hush up.) The usb_ethersubr module maintains a single queue of its own. When a packet is received in the USB callback routine, it's placed on this queue with usb_ether_input(). This routine also schedules a soft net interrupt with schednetisr(). The ISR routine then runs later, at splnet, outside of the USB callback/interrupt context, and passes the packet to ether_input(), hopefully in a safe manner. The reason this is implemented as a separate module is that there are a limited number of NETISRs that we can use, and snarfing one up for each driver that needs it is wasteful (there will be three once I get the CATC driver done). It also reduces code duplication to a certain small extent. Unfortunately, it also needs to be linked in with the usb.ko module in order for the USB ethernet drivers to share it. Also removed some uneeded includes from if_aue.c and if_kue.c Fix suggested by: peter Not rejected as a hairbrained idea by: n_hibma
2000-01-10 23:12:54 +00:00
if (status == USBD_NOT_STARTED || status == USBD_CANCELLED)
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
return;
printf("aue%d: usb error on rx: %s\n", sc->aue_unit,
usbd_errstr(status));
if (status == USBD_STALLED)
usbd_clear_endpoint_stall(sc->aue_ep[AUE_ENDPT_RX]);
goto done;
}
usbd_get_xfer_status(xfer, NULL, NULL, &total_len, NULL);
/*
* See if we've already accumulated some data from
* a previous transfer.
*/
if (c->aue_accum) {
total_len += c->aue_accum;
c->aue_accum = 0;
}
if (total_len <= 4 + ETHER_CRC_LEN) {
ifp->if_ierrors++;
goto done;
}
m = c->aue_mbuf;
bcopy(mtod(m, char *) + total_len - 4, (char *)&r, sizeof(r));
/* Turn off all the non-error bits in the rx status word. */
r.aue_rxstat &= AUE_RXSTAT_MASK;
/*
* Check to see if this is just the first chunk of a
* split transfer. We really need a more reliable way
* to detect this.
*/
if (r.aue_pktlen != total_len && total_len == AUE_CUTOFF) {
c->aue_accum = AUE_CUTOFF;
usbd_setup_xfer(xfer, sc->aue_ep[AUE_ENDPT_RX],
c, mtod(c->aue_mbuf, char *) + AUE_CUTOFF,
AUE_CUTOFF, USBD_SHORT_XFER_OK,
USBD_NO_TIMEOUT, aue_rxeof);
usbd_transfer(xfer);
return;
}
if (r.aue_rxstat) {
ifp->if_ierrors++;
goto done;
}
/* No errors; receive the packet. */
total_len -= (4 + ETHER_CRC_LEN);
ifp->if_ipackets++;
m->m_pkthdr.rcvif = (struct ifnet *)&aue_qdat;
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
m->m_pkthdr.len = m->m_len = total_len;
Attempt to fix a problem with receiving packets on USB ethernet interfaces. Packets are received inside USB bulk transfer callbacks, which run at splusb() (actually splbio()). The packet input queues are meant to be manipulated at splimp(). However the locking apparently breaks down under certain circumstances and the input queues can get trampled. There's a similar problem with if_ppp, which is driven by hardware/tty interrupts from the serial driver, but which must also manipulate the packet input queues at splimp(). The fix there is to use a netisr, and that's the fix I used here. (I can hear you groaning back there. Hush up.) The usb_ethersubr module maintains a single queue of its own. When a packet is received in the USB callback routine, it's placed on this queue with usb_ether_input(). This routine also schedules a soft net interrupt with schednetisr(). The ISR routine then runs later, at splnet, outside of the USB callback/interrupt context, and passes the packet to ether_input(), hopefully in a safe manner. The reason this is implemented as a separate module is that there are a limited number of NETISRs that we can use, and snarfing one up for each driver that needs it is wasteful (there will be three once I get the CATC driver done). It also reduces code duplication to a certain small extent. Unfortunately, it also needs to be linked in with the usb.ko module in order for the USB ethernet drivers to share it. Also removed some uneeded includes from if_aue.c and if_kue.c Fix suggested by: peter Not rejected as a hairbrained idea by: n_hibma
2000-01-10 23:12:54 +00:00
/* Put the packet on the special USB input queue. */
usb_ether_input(m);
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
2000-01-14 00:49:28 +00:00
return;
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
done:
2000-01-14 00:49:28 +00:00
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
/* Setup new transfer. */
usbd_setup_xfer(xfer, sc->aue_ep[AUE_ENDPT_RX],
c, mtod(c->aue_mbuf, char *), AUE_CUTOFF, USBD_SHORT_XFER_OK,
USBD_NO_TIMEOUT, aue_rxeof);
usbd_transfer(xfer);
2000-01-14 00:49:28 +00:00
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
return;
}
/*
* A frame was downloaded to the chip. It's safe for us to clean up
* the list buffers.
*/
static void aue_txeof(xfer, priv, status)
usbd_xfer_handle xfer;
usbd_private_handle priv;
usbd_status status;
{
struct aue_softc *sc;
struct aue_chain *c;
struct ifnet *ifp;
usbd_status err;
int s;
s = splimp();
c = priv;
sc = c->aue_sc;
ifp = &sc->arpcom.ac_if;
if (status != USBD_NORMAL_COMPLETION) {
if (status == USBD_NOT_STARTED || status == USBD_CANCELLED) {
splx(s);
return;
}
printf("aue%d: usb error on tx: %s\n", sc->aue_unit,
usbd_errstr(status));
if (status == USBD_STALLED)
usbd_clear_endpoint_stall(sc->aue_ep[AUE_ENDPT_TX]);
splx(s);
return;
}
ifp->if_timer = 0;
ifp->if_flags &= ~IFF_OACTIVE;
usbd_get_xfer_status(c->aue_xfer, NULL, NULL, NULL, &err);
c->aue_mbuf->m_pkthdr.rcvif = ifp;
usb_tx_done(c->aue_mbuf);
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
c->aue_mbuf = NULL;
if (err)
ifp->if_oerrors++;
else
ifp->if_opackets++;
splx(s);
return;
}
static void aue_tick(xsc)
void *xsc;
{
struct aue_softc *sc;
struct ifnet *ifp;
struct mii_data *mii;
int s;
s = splimp();
sc = xsc;
if (sc == NULL) {
splx(s);
return;
}
ifp = &sc->arpcom.ac_if;
mii = device_get_softc(sc->aue_miibus);
if (mii == NULL) {
splx(s);
return;
}
mii_tick(mii);
if (!sc->aue_link) {
mii_pollstat(mii);
if (mii->mii_media_status & IFM_ACTIVE &&
IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE)
sc->aue_link++;
if (ifp->if_snd.ifq_head != NULL)
aue_start(ifp);
}
sc->aue_stat_ch = timeout(aue_tick, sc, hz);
splx(s);
return;
}
static int aue_encap(sc, m, idx)
struct aue_softc *sc;
struct mbuf *m;
int idx;
{
int total_len;
struct aue_chain *c;
usbd_status err;
c = &sc->aue_cdata.aue_tx_chain[idx];
/*
* Copy the mbuf data into a contiguous buffer, leaving two
* bytes at the beginning to hold the frame length.
*/
m_copydata(m, 0, m->m_pkthdr.len, c->aue_buf + 2);
c->aue_mbuf = m;
/*
* XXX I don't understand why, but transfers that
* are exactly a multiple of 64 bytes in size don't
* work. I'm not sure why. If we detect such a
* transfer, we pad it out by one extra byte.
*/
total_len = m->m_pkthdr.len + 2;
if (!(total_len % 64))
total_len++;
/*
* The ADMtek documentation says that the packet length is
* supposed to be specified in the first two bytes of the
* transfer, however it actually seems to ignore this info
* and base the frame size on the bulk transfer length.
*/
c->aue_buf[0] = (u_int8_t)m->m_pkthdr.len;
c->aue_buf[1] = (u_int8_t)(m->m_pkthdr.len >> 8);
This commit adds device driver support for the ADMtek AN986 Pegasus USB ethernet chip. Adapters that use this chip include the LinkSys USB100TX. There are a few others, but I'm not certain of their availability in the U.S. I used an ADMtek eval board for development. Note that while the ADMtek chip is a 100Mbps device, you can't really get 100Mbps speeds over USB. Regardless, this driver uses miibus to allow speed and duplex mode selection as well as autonegotiation. Building and kldloading the driver as a module is also supported. Note that in order to make this driver work, I had to make what some may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer() function will use tsleep() for synchronous transfers that don't complete right away. This is a problem since there are times when we need to do sync transfers from an interrupt context (i.e. when reading registers from the MAC via the control endpoint), where tsleep() us a no-no. My hack allows the driver to have the code poll for transfer completion subject to the xfer->timeout timeout rather that calling tsleep(). This hack is controlled by a quirk entry and is only enabled for the ADMtek device. Now, I'm sure there are a few of you out there ready to jump on me and suggest some other approach that doesn't involve a busy wait. The only solution that might work is to handle the interrupts in a kernel thread, where you may have something resembling a process context that makes it okay to tsleep(). This is lovely, except we don't have any mechanism like that now, and I'm not about to implement such a thing myself since it's beyond the scope of driver development. (Translation: I'll be damned if I know how to do it.) If FreeBSD ever aquires such a mechanism, I'll be glad to revisit the driver to take advantage of it. In the meantime, I settled for what I perceived to be the solution that involved the least amount of code changes. In general, the hit is pretty light. Also note that my only USB test box has a UHCI controller: I haven't I don't have a machine with an OHCI controller available. Highlights: - Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part. - Updated usbdevs and regenerated generated files - Updated HARDWARE.TXT and RELNOTES.TXT files - Updated sysinstall/device.c and userconfig.c - Updated kernel configs -- device aue0 is commented out by default - Updated /sys/conf/files - Added new kld module directory
1999-12-28 02:01:18 +00:00
usbd_setup_xfer(c->aue_xfer, sc->aue_ep[AUE_ENDPT_TX],
c, c->aue_buf, total_len, 0, 10000, aue_txeof);
/* Transmit */
err = usbd_transfer(c->aue_xfer);
if (err != USBD_IN_PROGRESS) {
aue_stop(sc);
return(EIO);
}
sc->aue_cdata.aue_tx_cnt++;
return(0);
}
static void aue_start(ifp)
struct ifnet *ifp;
{
struct aue_softc *sc;
struct mbuf *m_head = NULL;
sc = ifp->if_softc;
if (!sc->aue_link)
return;
if (ifp->if_flags & IFF_OACTIVE)
return;
IF_DEQUEUE(&ifp->if_snd, m_head);
if (m_head == NULL)
return;
if (aue_encap(sc, m_head, 0)) {
IF_PREPEND(&ifp->if_snd, m_head);
ifp->if_flags |= IFF_OACTIVE;
return;
}
/*
* If there's a BPF listener, bounce a copy of this frame
* to him.
*/
if (ifp->if_bpf)
bpf_mtap(ifp, m_head);
ifp->if_flags |= IFF_OACTIVE;
/*
* Set a timeout in case the chip goes out to lunch.
*/
ifp->if_timer = 5;
return;
}
static void aue_init(xsc)
void *xsc;
{
struct aue_softc *sc = xsc;
struct ifnet *ifp = &sc->arpcom.ac_if;
struct mii_data *mii;
struct aue_chain *c;
usbd_status err;
int i, s;
if (ifp->if_flags & IFF_RUNNING)
return;
s = splimp();
/*
* Cancel pending I/O and free all RX/TX buffers.
*/
aue_reset(sc);
mii = device_get_softc(sc->aue_miibus);
/* Set MAC address */
for (i = 0; i < ETHER_ADDR_LEN; i++)
csr_write_1(sc, AUE_PAR0 + i, sc->arpcom.ac_enaddr[i]);
/* If we want promiscuous mode, set the allframes bit. */
if (ifp->if_flags & IFF_PROMISC) {
AUE_SETBIT(sc, AUE_CTL2, AUE_CTL2_RX_PROMISC);
} else {
AUE_CLRBIT(sc, AUE_CTL2, AUE_CTL2_RX_PROMISC);
}
/* Init TX ring. */
if (aue_tx_list_init(sc) == ENOBUFS) {
printf("aue%d: tx list init failed\n", sc->aue_unit);
splx(s);
return;
}
/* Init RX ring. */
if (aue_rx_list_init(sc) == ENOBUFS) {
printf("aue%d: rx list init failed\n", sc->aue_unit);
splx(s);
return;
}
sc->aue_cdata.aue_ibuf = malloc(AUE_INTR_PKTLEN, M_USBDEV, M_NOWAIT);
/* Load the multicast filter. */
aue_setmulti(sc);
/* Enable RX and TX */
csr_write_1(sc, AUE_CTL0, AUE_CTL0_RXSTAT_APPEND|AUE_CTL0_RX_ENB);
AUE_SETBIT(sc, AUE_CTL0, AUE_CTL0_TX_ENB);
AUE_SETBIT(sc, AUE_CTL2, AUE_CTL2_EP3_CLR);
mii_mediachg(mii);
/* Open RX and TX pipes. */
err = usbd_open_pipe(sc->aue_iface, sc->aue_ed[AUE_ENDPT_RX],
USBD_EXCLUSIVE_USE, &sc->aue_ep[AUE_ENDPT_RX]);
if (err) {
printf("aue%d: open rx pipe failed: %s\n",
sc->aue_unit, usbd_errstr(err));
splx(s);
return;
}
usbd_open_pipe(sc->aue_iface, sc->aue_ed[AUE_ENDPT_TX],
USBD_EXCLUSIVE_USE, &sc->aue_ep[AUE_ENDPT_TX]);
if (err) {
printf("aue%d: open tx pipe failed: %s\n",
sc->aue_unit, usbd_errstr(err));
splx(s);
return;
}
err = usbd_open_pipe_intr(sc->aue_iface, sc->aue_ed[AUE_ENDPT_INTR],
USBD_SHORT_XFER_OK, &sc->aue_ep[AUE_ENDPT_INTR], sc,
sc->aue_cdata.aue_ibuf, AUE_INTR_PKTLEN, aue_intr);
if (err) {
printf("aue%d: open intr pipe failed: %s\n",
sc->aue_unit, usbd_errstr(err));
splx(s);
return;
}
/* Start up the receive pipe. */
for (i = 0; i < AUE_RX_LIST_CNT; i++) {
c = &sc->aue_cdata.aue_rx_chain[i];
usbd_setup_xfer(c->aue_xfer, sc->aue_ep[AUE_ENDPT_RX],
c, mtod(c->aue_mbuf, char *), AUE_CUTOFF,
USBD_SHORT_XFER_OK, USBD_NO_TIMEOUT, aue_rxeof);
usbd_transfer(c->aue_xfer);
}
ifp->if_flags |= IFF_RUNNING;
ifp->if_flags &= ~IFF_OACTIVE;
(void)splx(s);
sc->aue_stat_ch = timeout(aue_tick, sc, hz);
return;
}
/*
* Set media options.
*/
static int aue_ifmedia_upd(ifp)
struct ifnet *ifp;
{
struct aue_softc *sc;
struct mii_data *mii;
sc = ifp->if_softc;
mii = device_get_softc(sc->aue_miibus);
sc->aue_link = 0;
if (mii->mii_instance) {
struct mii_softc *miisc;
for (miisc = LIST_FIRST(&mii->mii_phys); miisc != NULL;
miisc = LIST_NEXT(miisc, mii_list))
mii_phy_reset(miisc);
}
mii_mediachg(mii);
return(0);
}
/*
* Report current media status.
*/
static void aue_ifmedia_sts(ifp, ifmr)
struct ifnet *ifp;
struct ifmediareq *ifmr;
{
struct aue_softc *sc;
struct mii_data *mii;
sc = ifp->if_softc;
mii = device_get_softc(sc->aue_miibus);
mii_pollstat(mii);
ifmr->ifm_active = mii->mii_media_active;
ifmr->ifm_status = mii->mii_media_status;
return;
}
static int aue_ioctl(ifp, command, data)
struct ifnet *ifp;
u_long command;
caddr_t data;
{
struct aue_softc *sc = ifp->if_softc;
struct ifreq *ifr = (struct ifreq *) data;
struct mii_data *mii;
int s, error = 0;
s = splimp();
switch(command) {
case SIOCSIFADDR:
case SIOCGIFADDR:
case SIOCSIFMTU:
error = ether_ioctl(ifp, command, data);
break;
case SIOCSIFFLAGS:
if (ifp->if_flags & IFF_UP) {
if (ifp->if_flags & IFF_RUNNING &&
ifp->if_flags & IFF_PROMISC &&
!(sc->aue_if_flags & IFF_PROMISC)) {
AUE_SETBIT(sc, AUE_CTL2, AUE_CTL2_RX_PROMISC);
} else if (ifp->if_flags & IFF_RUNNING &&
!(ifp->if_flags & IFF_PROMISC) &&
sc->aue_if_flags & IFF_PROMISC) {
AUE_CLRBIT(sc, AUE_CTL2, AUE_CTL2_RX_PROMISC);
} else if (!(ifp->if_flags & IFF_RUNNING))
aue_init(sc);
} else {
if (ifp->if_flags & IFF_RUNNING)
aue_stop(sc);
}
sc->aue_if_flags = ifp->if_flags;
error = 0;
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
aue_setmulti(sc);
error = 0;
break;
case SIOCGIFMEDIA:
case SIOCSIFMEDIA:
mii = device_get_softc(sc->aue_miibus);
error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
break;
default:
error = EINVAL;
break;
}
(void)splx(s);
return(error);
}
static void aue_watchdog(ifp)
struct ifnet *ifp;
{
struct aue_softc *sc;
sc = ifp->if_softc;
ifp->if_oerrors++;
printf("aue%d: watchdog timeout\n", sc->aue_unit);
/*
* The polling business is a kludge to avoid allowing the
* USB code to call tsleep() in usbd_delay_ms(), which will
* kill us since the watchdog routine is invoked from
* interrupt context.
*/
sc->aue_udev->bus->use_polling++;
aue_stop(sc);
aue_init(sc);
sc->aue_udev->bus->use_polling--;
if (ifp->if_snd.ifq_head != NULL)
aue_start(ifp);
return;
}
/*
* Stop the adapter and free any mbufs allocated to the
* RX and TX lists.
*/
static void aue_stop(sc)
struct aue_softc *sc;
{
usbd_status err;
struct ifnet *ifp;
int i;
ifp = &sc->arpcom.ac_if;
ifp->if_timer = 0;
csr_write_1(sc, AUE_CTL0, 0);
csr_write_1(sc, AUE_CTL1, 0);
aue_reset(sc);
untimeout(aue_tick, sc, sc->aue_stat_ch);
/* Stop transfers. */
if (sc->aue_ep[AUE_ENDPT_RX] != NULL) {
err = usbd_abort_pipe(sc->aue_ep[AUE_ENDPT_RX]);
if (err) {
printf("aue%d: abort rx pipe failed: %s\n",
sc->aue_unit, usbd_errstr(err));
}
err = usbd_close_pipe(sc->aue_ep[AUE_ENDPT_RX]);
if (err) {
printf("aue%d: close rx pipe failed: %s\n",
sc->aue_unit, usbd_errstr(err));
}
sc->aue_ep[AUE_ENDPT_RX] = NULL;
}
if (sc->aue_ep[AUE_ENDPT_TX] != NULL) {
err = usbd_abort_pipe(sc->aue_ep[AUE_ENDPT_TX]);
if (err) {
printf("aue%d: abort tx pipe failed: %s\n",
sc->aue_unit, usbd_errstr(err));
}
err = usbd_close_pipe(sc->aue_ep[AUE_ENDPT_TX]);
if (err) {
printf("aue%d: close tx pipe failed: %s\n",
sc->aue_unit, usbd_errstr(err));
}
sc->aue_ep[AUE_ENDPT_TX] = NULL;
}
if (sc->aue_ep[AUE_ENDPT_INTR] != NULL) {
err = usbd_abort_pipe(sc->aue_ep[AUE_ENDPT_INTR]);
if (err) {
printf("aue%d: abort intr pipe failed: %s\n",
sc->aue_unit, usbd_errstr(err));
}
err = usbd_close_pipe(sc->aue_ep[AUE_ENDPT_INTR]);
if (err) {
printf("aue%d: close intr pipe failed: %s\n",
sc->aue_unit, usbd_errstr(err));
}
sc->aue_ep[AUE_ENDPT_INTR] = NULL;
}
/* Free RX resources. */
for (i = 0; i < AUE_RX_LIST_CNT; i++) {
if (sc->aue_cdata.aue_rx_chain[i].aue_buf != NULL) {
free(sc->aue_cdata.aue_rx_chain[i].aue_buf, M_USBDEV);
sc->aue_cdata.aue_rx_chain[i].aue_buf = NULL;
}
if (sc->aue_cdata.aue_rx_chain[i].aue_mbuf != NULL) {
m_freem(sc->aue_cdata.aue_rx_chain[i].aue_mbuf);
sc->aue_cdata.aue_rx_chain[i].aue_mbuf = NULL;
}
if (sc->aue_cdata.aue_rx_chain[i].aue_xfer != NULL) {
usbd_free_xfer(sc->aue_cdata.aue_rx_chain[i].aue_xfer);
sc->aue_cdata.aue_rx_chain[i].aue_xfer = NULL;
}
}
/* Free TX resources. */
for (i = 0; i < AUE_TX_LIST_CNT; i++) {
if (sc->aue_cdata.aue_tx_chain[i].aue_buf != NULL) {
free(sc->aue_cdata.aue_tx_chain[i].aue_buf, M_USBDEV);
sc->aue_cdata.aue_tx_chain[i].aue_buf = NULL;
}
if (sc->aue_cdata.aue_tx_chain[i].aue_mbuf != NULL) {
m_freem(sc->aue_cdata.aue_tx_chain[i].aue_mbuf);
sc->aue_cdata.aue_tx_chain[i].aue_mbuf = NULL;
}
if (sc->aue_cdata.aue_tx_chain[i].aue_xfer != NULL) {
usbd_free_xfer(sc->aue_cdata.aue_tx_chain[i].aue_xfer);
sc->aue_cdata.aue_tx_chain[i].aue_xfer = NULL;
}
}
free(sc->aue_cdata.aue_ibuf, M_USBDEV);
sc->aue_cdata.aue_ibuf = NULL;
sc->aue_link = 0;
ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
return;
}
/*
* Stop all chip I/O so that the kernel's probe routines don't
* get confused by errant DMAs when rebooting.
*/
static void aue_shutdown(dev)
device_t dev;
{
struct aue_softc *sc;
sc = device_get_softc(dev);
aue_reset(sc);
aue_stop(sc);
return;
}